Stem Separation?

[John Vere]

Stem Separation has been a major part of a few of my important workflows since I started using it about 3 years ago.
I tested about 10 different apps back then.

The one that Audacity uses is by far the best quality of the “free” category. SpectraLayers is expensive but wins first place in my opinion. But for now I only can afford the Elements version that is just Vocals and the Music.

A couple of use cases that make my life better.
Creating midi backing tracks. And removing Mike bleed from live recordings.

How I now create an excellent backing track!

I play gigs as a solo using backing tracks Ive made myself since DrT on an Atari. They have always used midi but I export them to stereo wave for playback.
In the past this required starting from scratch and learning the bass and drum parts. It often required 2 days of work to learn the parts and finish a backing track.

Now I take the original version using YouTube and record it into Audacity. I stem separate and export the 4 Stems.
I drop them into my Daw and first extract a tempo map from the drum stem.

I then use Prism ( the best audio too midi app) to convert the bass stem into midi. Prism now has added drums so I convert the drum stem into midi drums. This is a new toy and simply amazing!
I now have a backing track that is almost identical to the original performance but it is midi not the audio. So Im technically not breaking copyright! Midi keeps the tracks consistent song to song. I can also change the key.

I don’t need the guitar or vocals. Now it takes about two hours at most. I might add a piano or organ. Because of the tempo map the midi can be quantized. Huge time saver as well as deadly accurate version of a cover tune. I am actually playing along with the original drummer and bass player!
So two different tools used, Audacity stem separation and Prism audio to Midi.

Another use case- Mike bleed from live performance!

I had these clients in my studio who insisted on recording their music live and using one LDC Mike. vocals and Acoustic guitar. Well that sounded great but then they asked me after the fact to make some pitch correction with Melodyne to the vocals!
I have Melodyne Studio and I even contacted Celemony. No you can’t do that!
But a person on the Sonar forum suggested I try SpectraLayers and use separation first. They asked for a copy of the track.
He sent back the two stems the vocals and guitar and I was impressed! It sounded clean.
I ended up purchasing it and it has paid for itself 10 times over now. The client were very pleased!

It separated the vocals and the guitar with very minimal artifacts at all. Audacity can’t do that.

I also have used stem separation on drum mikes and especially the overheads to remove the bleeding of guitar, bass and vocals.This really cleans things up.
And like the other two old farts I’ve used it to remix old four track cassette recordings. That’s a gas!
Im hoping for a Black Friday sale on the full version of SpectraLayers. I think it’s included in the full version of Cubase.

 

[Craig Anderton]

TBH I think the use cases described in this thread are very much edge cases rather than the mainstream use for stems

Perhaps the most mainstream application would be Karaoke - separate out the vocal, flip it out of phase, and bounce it with the original track.

 

[SwitchBack]

... or separate 'vocals' and 'the rest' and mute the vocals.

 

[John Vere]

I know that bands use it to learn new cover songs. Say you’re the drummer. Just turn the drums off or down and play along.
If the band picks the same version then they don’t even need to get together to rehearse.
No fighting over how the arrangement goes!

 

[Craig Anderton]

... or separate 'vocals' and 'the rest' and mute the vocals.

Try it both ways. TBH I haven't used the latest iteration of stem separation, but with the original process, sometimes it sounded better by leaving the original vocal in and doing the phase flip. My theory is that the difference was due to the separated vocal not being exactly the same as the original vocal.

 

[John Vere]

The thing is that even the worst stem separation tool will sound exactly like the original if you play all the tracks. It’s when you solo a track that the artifacts rear their ugly heads.
So simply remixing a song you might not hear any quality loss because all the pieces are still there.

The artifacts are what will make the solo stem unusable as the track will contain leakage from the other tracks.
One example! The more stereo reverb the original had the worst it will split into 4 or 5 parts. The reverb contains the whole mix and confused the algorithm. The better quality stem separation seems to be able to deal with this problem better.
The lower quality algorithms like in Mixcraft make everything sound underwater when soloed.

 

[Craig Anderton]

So here's an interesting legal question: What about a tribute band removing the parts they play from a recording by the band they're emulating, but keeping the parts they're not playing as a backing track?

I was rehearsing "I Fought the Law and the Law Won" for a Halloween project, and removed the vocal and guitar so I could play along while practicing. Then the above question occurred to me. Hmmm...don't clubs and restaurants pay a blanket licensing fee to ASCAP and BMI? Seems to me maybe tribute bands could get away with it. Or at least maybe they can until they're told they can't.

 

[John Vere]

Im not sure what stem separation app you use but all the ones I have tried the quality of the backing track created by removing the vocals and “Other” stems sounds terrible! It’s why I convert it into midi. Have you soloed the bass part?
But I’m sure there’s many lazy people who don’t care about the sound quality and might be doing exactly what you propose.
It’s like when you walk into a restaurant and there’s a duo using a 1995 Casio keyboard for midi. Wonderful!
But yes this is the fact about playing cover music. The venue is responsible for licensing not the band.
It was why we had elevator music in the early days in shopping malls and dentist offices. The music was all instrumental versions of popular songs played by musicians who were paid once.
The venue paid a low monthly fee for low quality background music.
But in Canada all bars who featured live bands were charged based on the capacity of their venue.
Many small bars in small towns were unaware of this and got away with it for years. When Napster started killing royalties SOCAN went on a massive witch hunt and started busting even small businesses if they walked in and heard that the background music was a CD or even the radio.
My wife worked in a Dentist office and they were fined for playing the FM radio without permission. There were forced to pay for a licensed Muzak system.
They now avoid that by simply streaming because you pay for it ,,therefore it is considered legal. Yet another example of musicians not getting paid thanks to the streaming system.

 

[SwitchBack]

Good question. I tend to use the 'is money made' criterium. Private rehearsals are fine but as soon as there's a paying audience (tickets or drinks or whatever) a fee is due. Works for me, but then no crowd = no inspector. Never had to play the ignorance card

 

[Ken Morgan]

Good question. I tend to use the 'is money made' criterium. Private rehearsals are fine but as soon as there's a paying audience (tickets or drinks or whatever) a fee is due. Works for me, but then no crowd = no inspector. Never had to play the ignorance card

Everything is so confusing these days. What constitutes fair use, using a separated stem from the original recording in one's performance, the 'money made' criteria (before or after taxes and expenses?) and so on?

In that same vein, what makes a musician? Classical folks can sight read almost anything and reproduce it as written with frightened accurate results, but so many times cannot vamp a single bar. Some can just RIP a solo, but freeze at the sight of a chart (hello Glen Campbell). Some are so gifted at computer tech they actually have an understanding of it all at a programmers level, while others can barely check email without stress but can play any gig in any genre with a 1 in a million talent level.

Only point: since there really are no universally accepted definitions, seems there can't be any universally accepted guidelines. Paraphrasing what Craig said, let your conscious be your guide.